Background:  Use of computerized observational databases to compare anti-retroviral medications has not been fully validated. Concordance of response between patients in 3 observational databases and 3 randomized clinical trials has been demonstrated, but that study compared treatment strategies (i.e. HAART vs dual therapy) and did not compare specific agents. To gain better insight into the use of databases for comparing anti-retroviral regimens, we used the Parkland HIV Database to model 3 randomized placebo-controlled studies of anti-retroviral therapy. The targeted studies were ACTG 384, the 2NN study, and ACTG 5095.  

Methods:  The Parkland HIV Database was queried for anti-retroviral-naïve patients meeting the entry criteria for each study. The primary endpoint for each study was the time to virologic failure as determined by Kaplan-Meier analysis. Significant co-variates for each analysis were performed using the Cox proportional hazards model.

Results:  To model the ACTG 384 study, 793 patients treated with ZDV + 3TC or D4T + 3TC plus either EFV or NLF were identified. Patients treated with ZDV + 3TC + EFV had the longest time to treatment failure (p <0.01), but there no significant difference between the other 3 combinations at 96 weeks. In the 2NN model, 200 patients treated with D4T + 3TC and either NVP or EFV. At 48 weeks, patients treated with EFV had a slightly better time to treatment failure, but this was not statistically significant (p = 0.21). At 204 weeks, however, there was a significantly worse time to treatment failure in the NVP arm (p <0.01). In the ACTG 5095 model, 447 patients treated with ZDV + 3TC + EFV or ZDV + 3TC + ABC were identified. Time to virologic failure was better in patients treated with ZDV + 3TC + EFV compared with ZDV + 3TC + ABC at 96 weeks. In this study EFV was superior to ABC in subgroups with subgroups with baseline HIV-1 RNA >100,000 copies/mL, HIV-1 RNA <100,000 copies/mL and in those with at least one HIV-1 RNA <400 copies/mL.

Conclusions:  In this proof-of-concept study, each cohort had similar results as its corresponding clinical trial. This suggests that observational data utilizing specific entry criteria and analyzing specific regimens may provide good concordance between cohort analyses and randomized clinical trials of anti-retroviral therapy. Further study between the relationship between randomized trials and observational studies in the treatment of HIV-1 infection are warranted.

Keywords: Obsevational Studies; Clinical Trials; Concordance